CN115310472A - Nuclear pulse peak sequence-based one-dimensional convolution neural network nuclide identification method - Google Patents
Nuclear pulse peak sequence-based one-dimensional convolution neural network nuclide identification method Download PDFInfo
- Publication number
- CN115310472A CN115310472A CN202110492704.5A CN202110492704A CN115310472A CN 115310472 A CN115310472 A CN 115310472A CN 202110492704 A CN202110492704 A CN 202110492704A CN 115310472 A CN115310472 A CN 115310472A
- Authority
- CN
- China
- Prior art keywords
- model
- source
- neural network
- data
- different
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01T—MEASUREMENT OF NUCLEAR OR X-RADIATION
- G01T1/00—Measuring X-radiation, gamma radiation, corpuscular radiation, or cosmic radiation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- High Energy & Nuclear Physics (AREA)
- Measurement Of Radiation (AREA)
Abstract
The invention discloses a nuclear pulse peak sequence data-based one-dimensional convolution neural network nuclide identification method capable of ensuring nuclide identification accuracy and improving nuclide identification speed. The nuclear pulse peak value sequence data-based one-dimensional convolution neural network nuclide identification method comprises the steps of firstly actually measuring a nuclear pulse signal through a detector, filtering by using discrete wavelet transform, and then extracting the time corresponding to the amplitude value of a pulse and the amplitude value to form sequence data; carrying out maximum and minimum normalization operation on the obtained sequence data, and then dividing a data set into a training set and a test set; and (3) building a one-dimensional convolutional neural network by using TensorFlow, training a neural network model by using a training set, and then testing by using a testing set to optimize model parameters. By adopting the nuclear pulse peak sequence data-based one-dimensional convolution neural network nuclide identification method, nuclide identification can be realized quickly and accurately.
Description
Technical Field
The invention relates to nuclear radiation signal measurement and analysis and nuclide identification, in particular to a one-dimensional convolution neural network nuclide identification method based on nuclear pulse peak data.
Background
It is well known that: nuclide identification is the key content of radionuclide detection, and is a necessary index of the radionuclide identification in the fields of environmental radioactivity measurement, nuclear emergency, radioactive substance management and control and the like, namely 'quick and accurate'. Nowadays, nuclear energy science is rapidly developed, and applied to various industries, in order to better cope with sudden nuclear accidents, a faster and more accurate nuclide identification method needs to be researched. The whole process of the traditional gamma energy spectrum analysis method is as follows: the detector detects pulse data of nuclear signals, the pulse data is processed by a plurality of energy spectrometers to obtain original gamma energy spectrum data, and the original gamma energy spectrum data is analyzed by background deduction, smoothing, peak searching and the like, on the basis of energy scales, corresponding characteristic energy is obtained through characteristic peak positions and then matched with a nuclide library, and a nuclide type corresponding to the characteristic energy is obtained. The rapid development of the neural network is applied to various social industries, and a plurality of neural network methods are generated to analyze energy spectrum information so as to realize nuclide identification, and the methods improve the speed of nuclide identification, but are similar to the traditional nuclide identification in nature and all use nuclide energy spectrum data.
Due to the influence of factors such as detector performance and background noise, the accuracy of nuclide identification is influenced by the analysis of each section data in the gamma energy spectrum analysis. The traditional energy spectrum identification method is complex in process and time-consuming, and cannot meet the requirement of fast nuclide identification. The existing method for identifying nuclides by using a neural network does not break through the limitation of energy spectrum data and is limited in identification speed.
Disclosure of Invention
The invention aims to provide a nuclide identification method based on a one-dimensional convolution neural network of nuclear pulse peak sequence data, which can identify the accuracy of nuclides and improve the speed of nuclide identification.
The technical scheme adopted by the invention for solving the technical problems is as follows: the nuclear pulse peak sequence-based one-dimensional convolution neural network nuclide identification method comprises the following steps:
s1, measuring radioactive substances by using a gamma radiation detector to obtain original nuclear pulse data;
s2, inputting original nuclear pulse data into a trained one-dimensional convolutional neural network nuclide identification model, and outputting an identification result; the one-dimensional convolution neural network nuclide identification model is established by the following steps:
s21, measuring single sources, mixed sources, different distances, different angles and different moving speeds according to actual detection conditions; the method comprises single-source different-probe-source distance different-measurement-time experiments, mixed-source different-distance experiments, different-angle experiments and moving-source experiments; measuring nuclear pulse signals of the radioactive source of each measuring point under each condition;
s22, filtering pulse noise by adopting a discrete wavelet transform method;
s23, extracting the nuclear pulse peak value of the detector and the time of the corresponding peak value point in t time of each measuring point to obtain sequence data of N pulse peak values; acquiring a series of kernel pulses, wherein N kernel pulses are acquired at each measuring point, and extracting a peak value for each kernel pulse to obtain sequence data containing N pulse peak values;
s24, dividing pulse peak values in the N sequence data obtained in the S23 into N/T characteristic data sets according to every T data, wherein the shape of each data set is as follows: [ N/T, T ];
s25, combining the obtained data of all the measuring points into a sequence data set: m;
s26, adopting a maximum-minimum Normalization (Min-Max Normalization) method to normalize the data, and obtaining the data by the following formula:
wherein: x is a radical of a fluorine atom max Is the maximum value of the data set M; x is the number of min Is the minimum value of the data set M; x is a radical of a fluorine atom i Sample points for data set M;
s27, dividing the data set M into a training set x _ train and a test set x _ test according to the proportion of 6;
s28, building a one-dimensional convolutional neural network Model by adopting Python programming and using a TensorFlow frame, and recording the Model as a Model;
s29, training a Model by using the training set data x _ train;
determining parameters of each layer in the one-dimensional convolutional neural network; the convolution kernels are kept consistent in the whole training process, and finally the weight and the deviation of each neuron are gradually determined in the full-connection layer to obtain a one-dimensional convolution neural network model;
s210, testing model identification precision by using a test set, and evaluating a test effect by using a confusion matrix;
precision P for class c c As shown in the following formula:
wherein, TP c Being a true example of class c, FP c False positive examples for class c;
the confusion matrix is shown in the following table, in which TP c (True Positive) is a True example of category c; FN (FN) device c (False Negative) is a False Negative of class c; FP c (False Positive) is a False Positive case of class c; TN (twisted nematic) c (True Negative) is a True Negative of class c;
confusion matrix table for category c
S211, storing the model parameters, returning to the step S28 to adjust the model parameters, training a new model again, circulating the steps S28-S210 for 10 times or more, selecting the model with the best recognition effect as the model finally applied to nuclide recognition, and storing the model;
s212, loading a model, and identifying nuclides;
after the model is trained, the model parameters are stored, nuclide identification is carried out, and data are directly sent into the model, so that an output result is obtained.
Specifically, the experiment of different measurement times of the single source and different probe distances comprises the following contents:
by using 137 Cs、 60 Co、 133 Ba、 152 Eu、 155 The Eu standard source is designed for experiments of different distances between a radioactive source and the end face of the detector, the distance from 10cm to 1m is planned, the experiment of different measurement time on each point position is carried out at a measurement point of every 10 cm;
the different distance experiments of the mixed source refer to that: combining single sources into a mixed source to carry out different distance experiments;
the different angle experiments refer to: performing single-source and mixed-source experiments at different angles at different vertical distances from the surface of the detector;
the mobile source experiment refers to: and aiming at different moving speeds and different moving paths, carrying out a measurement experiment of a moving source, wherein the moving source is a single source or a mixed source.
Further, in step S211, the model parameters are saved, the process returns to step S28 to adjust the model parameters, the new model is trained again, and the process is repeated 10 to 50 times in steps S28 to S210.
Specifically, in step S22, the discrete wavelet transform filters impulse noise; the method sequentially comprises the following steps:
a. selecting wavelet and wavelet decomposition levels, and calculating wavelet decomposition from an original signal to an nth layer; selecting 3 decomposition layers, namely decomposing the low-frequency coefficient and the high-frequency coefficient into 3 layers;
b: selecting a threshold value for each layer of high-frequency coefficient, and correcting the high-frequency coefficient; the correction function is of the form:
thr＝max(x j ) (2)
in the formula, thr is a threshold value; k is an empirical coefficient, k is more than or equal to 0 and less than or equal to 1, when k =0 is equivalent to a hard threshold function, when k = l is equivalent to a soft threshold function, k =0.5 is taken; x is a radical of a fluorine atom jt And η jt The tth high-frequency coefficients of the jth layer before and after correction are respectively obtained; sign is a sign function;
c. wavelet reconstruction of the signal is performed based on the low frequency coefficient of the nth layer and the modified high frequency coefficients from the 1 st layer to the nth layer.
The invention has the beneficial effects that: according to the kernel identification method of the one-dimensional convolutional neural network based on the nuclear pulse peak sequence data, the final model for identifying the nuclide is obtained by establishing the one-dimensional convolutional neural network model and training the one-dimensional convolutional neural network model through experimental data, so that the nuclide identification precision and identification speed can be improved in the application process, dependence on the spectral data of the radionuclide is eliminated, complex time-consuming processes such as peak searching, smooth energy spectrum and the like are not needed, and the nuclide is identified in one step in advance by using the peak sequence data.
Drawings
FIG. 1 is a flow chart of a nuclear pulse sequence data-based one-dimensional convolutional neural network nuclide identification method in an embodiment of the present invention;
FIG. 2 is a confusion matrix of actual test results according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of an experiment with different distances according to an embodiment of the present invention;
FIG. 4 is a schematic diagram illustrating different angle detection according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of an experiment of a mobile source according to an embodiment of the present invention.
Detailed Description
The invention is further illustrated with reference to the following figures and examples.
With reference to fig. 1 to 5, the nuclide identification method based on the one-dimensional convolutional neural network of the nuclear pulse sequence according to the present invention includes the following steps:
s1, measuring radioactive substances by using a gamma radiation detector to obtain original nuclear pulse data;
and S2, inputting the original nuclear pulse data into the trained one-dimensional convolutional neural network nuclide identification model, and outputting an identification result.
The establishment process of the one-dimensional convolution neural network nuclide identification model is as follows:
s21, according to the actual detection condition, measuring experiments of single sources, mixed sources, different distances, different angles and different moving speeds are carried out, and the specific design is as follows:
carrying out single-source different probing source distance and different measurement time experiments;
by using 137 Cs、 60 Co、 133 Ba、 152 Eu、 155 Eu, etc., designing experiments of different distances between a radioactive source and the end face of the detector, preliminarily planning the distance from 10cm to 1m, and measuring points at intervals of 10cm, as shown in fig. 3. And then carrying out experiments of different measurement time on each point.
Carrying out different distance experiments on the mixed source;
combining single sources into mixed sources for carrying out experiments at different distances, e.g. 137 Cs+ 60 Co、 137 Cs+ 60 Co+ 133 Ba、 137 Cs+ 60 Co+ 133 Ba+ 152 Eu、 137 Cs+ 60 Co+ 133 Ba+ 152 Eu+ 155 Eu, and the like.
At different vertical distances from the detector surface, single source and mixed source experiments at different angles (e.g. 15 ° apart) are performed, as shown in fig. 4;
moving source experiment;
the measurement experiment of the moving source is carried out according to different moving speeds and different moving paths, and is also divided into the single source and mixed source conditions, and the experimental schematic diagram is shown in fig. 5.
And acquiring basic data required by training the neural network model, namely pulse peak sequence data of the impulse detection source based on the experiment.
S22, filtering pulse noise by adopting a discrete wavelet transform method;
specifically, a, selecting wavelet and wavelet decomposition levels, and calculating wavelet decomposition from an original signal to an nth layer. The project adopts a Coiflet (coifN) wavelet system widely applied to signal processing; a large number of experiments show that the more the number of decomposition layers, the more the removed high-frequency information, but the real information contained in the low-frequency part is correspondingly reduced, and the phenomenon that the spectral peak is pressed down is shown on the denoised spectrum, so that the number of decomposition layers is selected to be 3, namely, the low-frequency coefficient and the high-frequency coefficient are decomposed into 3 layers.
b. And selecting a threshold value for each layer of high-frequency coefficient, and correcting the high-frequency coefficient. The correction function is of the form:
thr＝max(x j ) (2)
in the formula, thr is a threshold value; k is an empirical coefficient, k is more than or equal to 0 and less than or equal to 1, when k =0 is equivalent to a hard threshold function, when k = l is equivalent to a soft threshold function, k =0.5 is taken; x is a radical of a fluorine atom jt And η jt The tth high-frequency coefficients of the jth layer before and after correction are respectively obtained; sign is a symbol function.
c. Wavelet reconstruction of the signal is performed based on the low frequency coefficient of the nth layer and the modified high frequency coefficients from the 1 st layer to the nth layer.
S23, extracting the time of the nuclear pulse peak value of the detector and the corresponding peak value point in t time of each measuring point to obtain N series data;
s24, dividing pulse peak values in the N sequence data obtained in the S23 into every T data, and dividing N/T characteristic data sets, wherein the shape of each data set is as follows: [ N/T, T ].
S25, combining the obtained data of all the measuring points into a sequence data set: m;
s26, adopting a maximum-minimum Normalization (Min-Max Normalization) method to normalize the data, wherein the formula (4) is as follows:
wherein: x is the number of max Is the maximum value of the data set M; x is the number of min Is the minimum value of the data set M; x is the number of i Sample points for data set M;
s27, dividing the data set M into a training set x _ train and a test set x _ test according to the proportion of 6;
s28, building a one-dimensional convolutional neural network Model by adopting Python programming and using a TensorFlow frame, and recording the Model as a Model; the parameters of the Model are shown in Table 1. The convolutional neural network used in the method has a total of 10 layers, including a convolutional layer pooling layer and a full connection layer.
TABLE 1 parameters of layers of one-dimensional convolutional neural network
S29, training a Model by using training set data x _ train;
and in the training process, determining parameters of each layer in the one-dimensional convolutional neural network. The convolution kernels are kept consistent in the whole training process, and finally the weight and the deviation of each neuron are gradually determined in the full connection layer to obtain a one-dimensional convolution neural network model.
And S210, testing the model identification precision by using the test set, and evaluating the test effect by using the confusion matrix.
Precision P for class c c As shown in formula (5). The confusion matrix is shown in Table 2, where TP in Table 2 c (True Positive) is a True example of class c; FN (FN) device c (False Negative) is a False Negative of class c; FP c (False Positive) is a False Positive case of class c; TN (twisted nematic) c (True Negative) is a True Negative of class c.
Wherein, TP c Being a true example of class c, FP c Is a false positive example of class c.
TABLE 2 confusion matrix for Category c
S211, storing the model parameters, returning to the step S19 to adjust the model parameters, training a new model again, circulating the steps S28-S210 for 10 times, selecting the model with the best recognition effect as the model finally applied to nuclide recognition, and storing the model.
S212, loading a model and identifying nuclides. After the model is trained, the model parameters are stored, and then data are directly sent into the model when nuclide identification is carried out, so that an output result can be obtained quickly.
Example 1
The nuclide identification method of the one-dimensional convolution neural network based on the nuclear pulse sequence loads the trained model pair 137 Cs， 60 Co， 155 Eu- 22 Na， 137 Cs- 60 Co is tested, and the confusion matrix of the test result is shown in figure 3; the accuracy of the test for each nuclide is shown in table 3.
TABLE 3 test accuracy of model for each nuclide
| Test nuclide | 137 Cs | 60 Co | 155 Eu- 22 Na | 137 Cs- 60 Co |
| Accuracy of measurement | 100％ | 99.99％ | 99.98％ | 99.76％ |
Claims (4)
1. The one-dimensional convolution neural network nuclide identification method based on the kernel pulse peak sequence is characterized by comprising the following steps of:
s1, measuring radioactive substances by using a gamma radiation detector to obtain original nuclear pulse data;
s2, inputting original nuclear pulse data into a trained one-dimensional convolutional neural network nuclide identification model, and outputting an identification result; the one-dimensional convolution neural network nuclide identification model is established by the following steps:
s21, measuring single sources, mixed sources, different distances, different angles and different moving speeds according to actual detection conditions; the method comprises single-source different-probe-source distance different-measurement-time experiments, mixed-source different-distance experiments, different-angle experiments and moving-source experiments; measuring nuclear pulse signals of the radioactive source of each measuring point under each condition;
s22, filtering pulse noise by adopting a discrete wavelet transform method;
s23, extracting the peak value of the nuclear pulse of the detector and the time of the corresponding peak value point in t time of each measuring point to obtain sequence data of N peak values; acquiring a series of N nuclear pulses at each measuring point, extracting each nuclear pulse to obtain sequence data containing N pulse peak values;
s24, dividing pulse peak values in the N sequence data obtained in the S23 into every T data, and dividing N/T characteristic data sets, wherein the shape of each data set is as follows: [ N/T, T ];
s25, combining the obtained data of all the measuring points into a sequence data set: m;
s26, adopting a maximum-minimum Normalization (Min-Max Normalization) method to normalize the data, and obtaining the data by the following formula:
wherein: x is a radical of a fluorine atom max Is the maximum value of the data set M; x is the number of min Is the minimum value of the data set M; x is a radical of a fluorine atom i Sample points for data set M;
s27, dividing the data set M into a training set x _ train and a test set x _ test according to the proportion of 6;
s28, adopting Python programming, and building a one-dimensional convolutional neural network Model by using a TensorFlow frame, and recording the Model as a Model;
s29, training a Model by using training set data x _ train;
determining parameters of each layer in the one-dimensional convolutional neural network; the convolution kernels are kept consistent in the whole training process, and finally the weight and the deviation of each neuron are gradually determined in the full connection layer to obtain a one-dimensional convolution neural network model;
s210, testing model identification precision by using a test set, and evaluating a test effect by using a confusion matrix;
precision P for class c c As shown in the following formula:
wherein, TP c Being a true example of class c, FP c False positive examples for class c;
the confusion matrix is shown in the following table, in which TP c (True Positive) of the category cTrue and true cases; FN (FN) c (False Negative) is a False Negative of class c; FP c (False Positive) is a False Positive case of class c; TN (twisted nematic) motor c (True Negative) is a True Negative of class c;
confusion matrix table for category c
S211, storing model parameters, returning to the step S28 to adjust the model parameters, training a new model again, circulating the steps S28-S210 for 10 times or more than 10 times, selecting the model with the best recognition effect as the model finally applied to nuclide recognition, and storing the model;
s212, loading a model, and identifying nuclides;
after the model is trained, the model parameters are stored, nuclide identification is carried out, and data are directly sent into the model, so that an output result is obtained.
2. The nuclide identification method for a one-dimensional convolutional neural network based on a kernel pulse peak sequence as set forth in claim 1, characterized in that:
the experiment of the single source and different probe source distances and different measuring times comprises the following contents:
by using 137 Cs、 60 Co、 133 Ba、 152 Eu、 155 The Eu standard source is used for designing experiments of different distances from a radioactive source to the end face of the detector, wherein the distance from 10cm to 1m is planned, the experiments of different measurement time on each point position are carried out at intervals of 10 cm;
the different distance experiments of the mixed source refer to that: combining single sources into a mixed source to carry out different distance experiments;
the different angle experiment refers to: performing single-source and mixed-source experiments at different angles at different vertical distances from the surface of the detector;
the mobile source experiment refers to: and aiming at different moving speeds and different moving paths, carrying out a measurement experiment of a moving source, wherein the moving source is a single source or a mixed source.
3. The nuclide identification method for a one-dimensional convolutional neural network based on a kernel pulse peak sequence as set forth in claim 2, characterized in that: the model parameters are saved in step S211, the model parameters are adjusted in step S28, the new model is trained again, and the process is repeated 10 to 50 times in steps S28 to S210.
4. The nuclide identification method for a one-dimensional convolutional neural network based on a kernel pulse peak sequence as set forth in claim 3, characterized in that: in step S22, the discrete wavelet transform filters impulse noise; the method sequentially comprises the following steps:
a. selecting wavelet and wavelet decomposition levels, and calculating wavelet decomposition from an original signal to an nth layer; selecting 3 decomposition layers, namely decomposing the low-frequency coefficient and the high-frequency coefficient into 3 layers;
b: selecting a threshold value for each layer of high-frequency coefficient, and correcting the high-frequency coefficient; the correction function is of the form:
thr＝max(x j ) (2)
in the formula, thr is a threshold value; k is an empirical coefficient, k is more than or equal to 0 and less than or equal to 1, when k =0 is equivalent to a hard threshold function, when k = l is equivalent to a soft threshold function, k =0.5 is taken; x is the number of jt And η jt The tth high-frequency coefficients of the jth layer before and after correction are respectively obtained; sign is a sign function;
c. wavelet reconstruction of the signal is performed based on the low frequency coefficient of the nth layer and the modified high frequency coefficients from the 1 st layer to the nth layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110492704.5A CN115310472A (en) | 2021-05-07 | 2021-05-07 | Nuclear pulse peak sequence-based one-dimensional convolution neural network nuclide identification method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110492704.5A CN115310472A (en) | 2021-05-07 | 2021-05-07 | Nuclear pulse peak sequence-based one-dimensional convolution neural network nuclide identification method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115310472A true CN115310472A (en) | 2022-11-08 |
Family
ID=83853585
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110492704.5A Pending CN115310472A (en) | 2021-05-07 | 2021-05-07 | Nuclear pulse peak sequence-based one-dimensional convolution neural network nuclide identification method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115310472A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115856987A (en) * | 2023-02-28 | 2023-03-28 | 西南科技大学 | Nuclear pulse signal and noise signal discrimination method under complex environment |
-
2021
- 2021-05-07 CN CN202110492704.5A patent/CN115310472A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115856987A (en) * | 2023-02-28 | 2023-03-28 | 西南科技大学 | Nuclear pulse signal and noise signal discrimination method under complex environment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107229787B (en) | Gamma energy spectrum analysis method based on approximation coefficient and deep learning | |
| CN108983279B (en) | Low-background spectrum resolving method based on sodium iodide detector | |
| CN109063741A (en) | A kind of energy spectrum analysis method based on hibert curve transformation and deep learning | |
| CN101632011A (en) | The advanced pattern recognition systems that is used for spectral analysis | |
| CN107632010B (en) | Method for quantifying steel sample by combining laser-induced breakdown spectroscopy | |
| CN113687433A (en) | Bi-LSTM-based magnetotelluric signal denoising method and system | |
| CN107390259A (en) | A kind of nuclide identification method based on SVD and SVM | |
| CN114925615A (en) | Radioactive leakage source positioning method and system | |
| CN114237046A (en) | Partial discharge pattern recognition method based on SIFT data feature extraction algorithm and BP neural network model | |
| CN115310472A (en) | Nuclear pulse peak sequence-based one-dimensional convolution neural network nuclide identification method | |
| CN115034254A (en) | Nuclide identification method based on HHT (Hilbert-Huang transform) frequency band energy features and convolutional neural network | |
| CN112232343B (en) | Grain mildew grain identification neural network and identification method | |
| CN113536894A (en) | Defect detecting method, defect detecting apparatus, and computer-readable storage medium | |
| CN116952586A (en) | Axle fatigue crack identification method based on 1D-DCNN | |
| Shaochuan | Bayesian multiple changepoint detection for stochastic models in continuous time | |
| CN117216522A (en) | Bearing fault type diagnosis method, device and equipment based on machine learning | |
| CN103455986A (en) | Random noise point detection algorithm based on fractional-order differential gradient | |
| CN111191502A (en) | Stick-slip and jump drill abnormal working condition identification method based on drill string vibration signal | |
| CN113818867B (en) | Method, system, medium, equipment and application for constructing pseudo capillary pressure curve | |
| CN111008356B (en) | Gamma energy spectrum set analysis method for deducting background based on WTS VD algorithm | |
| CN107590824B (en) | Rock particle identification and displacement tracking method based on three-dimensional image processing technology | |
| CN113239075A (en) | Construction data self-checking method and system | |
| JP2016186489A (en) | Spectrometric analysis method and related device | |
| CN114329333B (en) | NASDV (network-independent data decomposition) denoising effective singular value number determining method for aviation gamma energy spectrum data | |
| Tang et al. | Research on tool wear prediction based on deep residual network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |